Widespread vegetation changes have been evidenced by satellite-observed long-term trends over decades in vegetation indices (VIs). However, many issues can affect the derived VIs trends, among which the inherent difference between VIs calculated from the same input reflectance has not been investigated. Here, we compared global long-term trends in six widely used RED-NIR (near-infrared)-based VIs calculated from the MODIS nadir bidirectional reflectance distribution function (BRDF) adjusted product (MCD43A4) during 2000-2023, including normalized difference vegetation index (NDVI), kernel NDVI (kNDVI), 2-band enhanced vegetation index (EVI2), near-infrared reflectance of vegetation (NIRv), difference vegetation index (DVI), and plant phenology index (PPI). We identified two distinct groups of VIs, i.e., (1) NDVI and kNDVI, and (2) EVI2, NIRv, DVI, and PPI, which shared similar trends within the group but showed significant directional differences between groups in 17.4% of the studied area. Only 20.5% of the global land surface showed consistent trends. Based on the radiation transfer model and remote sensing observations, we demonstrated that the two groups of VIs differed in their sensitivities to RED and NIR reflectance. These differences lead to inconsistent long-term trends arising from variations in vegetation type, mixed pixel effects, saturation, and asynchronous changes in vegetation chlorophyll content and structural attributes. Comparisons with ground-observed leaf area index (LAI), flux tower gross primary productivity (GPP), and PhenoCam green chromatic coordinate (GCC) further revealed that the EVI2, NIRv, DVI, and PPI trends corresponded more closely with LAI and GPP trends, whereas the NDVI and kNDVI trends were more strongly associated with GCC trends. Our results highlight that long-term vegetation trends derived from different RED-NIR-based VIs must be interpreted by considering their intrinsic sensitivities to biophysical properties, which is essential for reliable assessments of vegetation dynamics.

Forest cover has expanded across tropical and subtropical Asia in recent decades, but area alone is an inadequate metric because it does not capture forest structure, which is critical for supporting key ecosystem functions and services, including biodiversity. Here, we apply sub-meter resolution satellite imagery to characterize the complexity of forests from a horizontal perspective by quantifying the diversity of tree crown sizes and their spatial arrangement across India, Southeast Asia, and southern China. We reveal large mismatches between reported forest area and the complexity of the forests. While overall, the majority of the forests still have a relatively high complexity, a recent change towards low complexity forests is observed. India, despite its large forest extent, shows a particularly low forest complexity, with more than half of its forests classified as low-complexity. Analysing forests affected by gain or loss in area between 2000 and 2024, we find that approximately three quarters of newly established forests are of low structural complexity, while high-complexity forests, concentrated in Myanmar, Laos, Cambodia, and Indonesia, continue to disappear. Our results highlight the need to move beyond forest cover statistics toward quality-based indicators for monitoring in the support of conservation policy.

Reed wetlands are key to the productivity of shallow lakes, and their condition is tightly governed by water level variability. Using long-term satellite observations, we provide the first analysis linking hydrology and reed vitality at Lake Neusiedl, a major climate sensitive wetland system in the Pannonian Basin. We assembled a 40-year record (1985–2025) of Landsat derived Enhanced Vegetation Index (EVI) and related it to in-situ measurements of surface water levels to quantify the impact of extreme events. The analyses of hydrological droughts in 1990, 2003 and 2022 (a record low) reveal a decline in plant vitality during periods of low water. Temporary drawdowns are expected to enhance vigour through sediment oxidation and litter decomposition, producing vegetation rebounds as the severity of the drought alleviates. A significant increase in productivity was observed following the 2022 event, but not after the 1990 and 2003 events. The analysis shows that reed vitality is not a continuous linear function of water availability or inundation level. Instead, it follows a pulse-response dynamic, where rejuvenation occurs only when critical thresholds of exposure depth and duration are exceeded, enabling oxidation of accumulated litter and sediment organic matter driving long-term die-back. The results provide a quantitative basis for adaptive water level management at Lake Neusiedl and demonstrate how long-term satellite monitoring can guide reed conservation in shallow lake wetlands more broadly.

A lack of historical land-use data hinders the appropriate assessment of ecosystem restoration potential. Here we integrate pollen and phytolith analysis, isotopic dating, historical records, and high-resolution satellite imagery to reconstruct five centuries of land-use and vegetation change in the karst landscapes of Southwest China. We show that prior to the 18th century, dense forests dominated the region under minimal human influence. However, the introduction of maize cultivation, rapid population growth, and extensive deforestation triggered a persistent shift toward open landscapes. This transformation is evidenced by a marked rise in pioneer fern spores ( 11

Tropical rainforests have complex responses to seasonal climatic variations. Compared to the Amazon and Asian rainforests, the Congo rainforest exhibits a more widespread bimodal seasonal pattern, yet it remains understudied. Here, we use three independent satellite-based vegetation indices to investigate the seasonal variations of the Congo rainforest, including solar-induced chlorophyll fluorescence (SIF), the two-band enhanced vegetation index (EVI2), and vegetation optical depth (VOD), which represent vegetation canopy photosynthesis, greenness, and water content, respectively. We find widespread asynchronous bimodal seasonality among the three vegetation indices, suggesting alternating physiological and structural variations of the Congo rainforest. These bimodal seasonality shifts are driven by temporally diverse environmental factors. The variations in photosynthesis are predominantly driven by temperature from January to June of the first growing season, and by precipitation and temperature from July to December of the second growing season. The bimodal seasonal pattern of vegetation greenness is primarily constrained by precipitation across most of the region. In contrast, variations in vegetation water content are closely related to terrestrial water storage, with a temporal lag of 1 to 2 months. These findings highlight the intertwined seasonality of the vegetation canopy traits and their environmental controls in the Congo rainforest, providing an improved understanding of the vegetation-climate feedback for predicting rainforest responses to climate change.

Counting individual trees is a fundamental task for environmental monitoring, yet remains largely unexplored with satellite imagery. At these resolutions, isolated trees may still be identifiable, but crown boundaries become ambiguous in dense forests, making the notion of an individual tree inherently ill-defined. Moreover, large-scale manual annotations of individual trees are prohibitively expensive. While scalable supervision can be derived from airborne LiDAR, the resulting annotations are noisy and difficult to exploit effectively. We address these challenges by formulating tree counting as a spatial density matching problem supervised through Unbalanced Optimal Transport. This formulation naturally accommodates both precise localization of isolate trees and robust density estimation in dense forests. We further introduce a self-correction mechanism that leverages transport residuals to progressively refine noisy supervision during training. We evaluate our approach on TinyTrees, a new benchmark spanning three continents and three satellite sensors, comprising over 215 million tree annotations (including 773K manually verified instances) across 23,000 sq.km. Our method consistently outperforms detection-based, regression-based, and transport-based distribution-matching baselines, demonstrating the effectiveness of unbalanced transport and reliability-aware supervision for large-scale tree counting from satellite imagery. Code, data and models are available at https://github.com/dgominski/treematch.

Forest disturbances are rising globally due to climate change and land-use pressures, weakening the terrestrial carbon sink. However, declines in aboveground biomass from tree cover loss remain poorly quantified, limiting our understanding of the role of disturbances in global carbon dynamics. Here we present a spatially explicit estimate of aboveground biomass losses from natural disturbances and harvest across Europe’s 216 million hectares of forests, using satellite remote sensing. From 1985 to 2023, gross aboveground biomass losses totalled 6.5 ± 0.8 Pg. Of these, 18% were caused by high-severity natural disturbances, whereas stand-replacing harvests accounted for 82%. From 2018 onward, aboveground biomass losses increased by 46% reaching annual values unprecedented in the preceding four decades. This acceleration coincided with high natural disturbance activity in biomass-rich temperate forests. After 2018, the sensitivity of aboveground biomass loss to disturbance area increased substantially, suggesting that even small increases in natural disturbances can result in large losses. As climate-driven natural disturbances intensify, we thus expect sustained aboveground biomass losses across Europe, despite policies to enhance the forest carbon sink until 2030. Satellite observations reveal accelerating aboveground biomass loss across Europe’s forests since 2018, increasingly driven by natural disturbances. These trends weaken the forest carbon sink and challenge climate mitigation efforts through forest policy.

Larval source management is a key strategy to control the spread of Aedes-borne viral diseases including dengue, Zika, chikungunya, and yellow fever. However, locating potential larval habitats through traditional field methods is challenging and labor-intensive at scale. Here, we demonstrate a scalable, high-resolution drone imagery and supervised machine learning approach to map potential Aedes larval container habitats across Dar es Salaam, Tanzania, a dense urban environment with informal settlements. Local larval surveillance and existing literature confirm epidemiological relevance of buckets and jerry cans, tires, and water tanks as key potential habitats. Drone images revealed rooftop tires, a container type likely overlooked during ground surveillance. We trained a U-Net deep learning model on very-high-resolution drone imagery (3-5 cm resolution) which was manually annotated across 4.6 km2, and applied it to predict containers across 27.27 km2, spanning 20 neighborhoods. The model predicted over 135,000 containers with detection accuracies of 75% for water tanks, 72% for tires, and 54% for buckets. Bucket and tire densities were strongly and positively correlated with population density across neighborhoods, whereas water tank density was not, suggesting the distribution of these container types reflect distinct underlying drivers. This study reveals otherwise difficult-to-observe container types, highlights the abundance and spatial heterogeneity of potential Aedes larval habitats across Dar es Salaam, and demonstrates a scalable approach for improving detection of potential Aedes larval container habitats in a dense urban environment.

Accurate tree counts are crucial for global restoration efforts, given the need for progress tracking and effective evaluation of tree planting and survival. However, conventional methods often fail to accurately capture tree counts at large scales, particularly in landscapes outside forests. We developed a deep learning approach for estimating tree counts from satellite imagery, nationally, across Rwanda and Tanzania in East Africa in 2019. We used a 20-band composite image composed of bands from PlanetScope, Sentinel-1, and Sentinel-2 as input, and estimated a total count of about 5.2 billion trees across the 2 countries. Only about 48.8% of the estimated tree counts fall within the WorldCover “tree cover” class, with the remainder being distributed across nonforest landscapes. Our work provides information on tree counts at the national scale in Rwanda and Tanzania using low-cost and freely available satellite images, which can guide the planning and optimization of resource allocation for tree restoration efforts at national scales.

Abstract. Above-ground biomass (AGB) maps are essential for carbon accounting and sustainable land management, yet AGB for non-forest landscapes remains poorly accounted for in global datasets. Here, we make use of deep learning and high-resolution PlanetScope imagery to introduce the concept of AGB contribution maps, which are high-resolution AGB predictions that capture local patterns. These maps can be predicted at any resolution from 1 to 100 m, providing insights into the spatial features included in the coarser resolution AGB maps, being essential for mapping trees outside forests. Our method employs a weakly supervised hybrid framework that transfers information from an existing 100 m global AGB map to high‑resolution optical satellite imagery, enabling the interpretation of detailed spatial patterns. We demonstrate that our map achieves detailed and spatially consistent patterns of woody vegetation in African savanna landscapes comparable to UAV-based LiDAR. Aggregated AGB values are well aligned with independent in-situ measurements (r2 = 0.71, bias 1 %), which is contrary to the original coarse AGB map used for training (r2 = 0.17, bias 48 %), indicating the capability of our approach to refine the existing map towards a higher accuracy for estimating tree biomass outside forests. This suggests that our model has learned tree-level information that is not present in the original AGB training data, providing a framework to refine existing coarse-resolution AGB maps. The granular and multi-resolution results provide no contribution to global efforts in sustainable land management of non-forest landscapes at any preferred scale and resolution.

Trees are key components of the terrestrial biosphere, playing vital roles in ecosystem function, climate regulation, and the bioeconomy. However, large-scale monitoring of individual trees remains limited by inadequate modeling. Available global products have focused on binary tree cover or canopy height, which do not explicitly identify trees at the individual level. In this study, we present a deep learning approach for detecting large individual trees in 3-m-resolution PlanetScope imagery at a global scale. We simulate tree crowns with Gaussian kernels of scalable size, allowing the extraction of crown centers and the generation of binary tree cover maps. Training is based on billions of points automatically extracted from airborne light detection and ranging (LiDAR) data, enabling the model to successfully identify trees both inside and outside forests. We compare against existing tree cover maps and airborne LiDAR with state-of-the-art performance (fractional cover R2 = 0.81 against aerial LiDAR), report balanced detection metrics across biomes, and demonstrate how detection can be further improved through fine-tuning with manual labels. Our method offers a scalable framework for global, high-resolution tree monitoring and is adaptable to future satellite missions offering improved imagery.

Computer vision algorithms have been widely employed for characterizing visually interpretable forest and tree structures such as individual tree crowns from high-resolution optical imagery. However, the learning of more complex and integrated variables, including forest biomass, remains challenging. Consequently, above-ground biomass (AGB) is traditionally estimated from structural measures of trees, which require Light Detection and Ranging (LiDAR) data that is not always readily available. Here, we evaluate the capacity of convolutional neural networks (CNN) to interpret spatial semantic patterns in optical RGB images to directly estimate forest AGB. Trained with forest inventory plots, the CNN model shows evidence of learning via implicitly interpreting the composition of biomass at tree level, differing from traditional approaches reliant on conversions of aggregated parameters. We compare the proposed method against benchmark approaches for biomass estimation at plot and tree levels. We further propose a learnable allometric model based on smooth min-max networks, which allows for flexible adjustments of allometric relationships between field-measured biomass and remotely detected tree features, such as LiDAR-derived tree height. Our experiments demonstrate that the CNN model yields arguably high performance (R2 = 0.71) without tree height information, approaching a random forest model with access to tree height (R2 = 0.8). CNN's ability to infer biomass directly from optical high-resolution images, down to the level of single pixels of the imagery used, enables a new level of flexibility in spatial scaling. Furthermore, when optical images are available at a higher temporal frequency than LiDAR data, CNNs offer enhanced flexibility in temporal scaling.

Abstract This study investigates the application of Physics‐Informed Neural Operators (PINOs) for solving the two‐dimensional shallow water equations (2D SWE) in the context of flood modeling. Unlike Physics‐Informed Neural Networks (PINNs), which require retraining for each new initial or boundary condition (BC), PINOs learn the solution operator itself, enabling fast inference across a range of conditions without retraining. PINNs have moreover been shown to struggle with complex bed topography due to the difficulty of satisfying the well‐balanced property (Tian et al., 2025, https://doi.org/10.1029/2025wr040052; Dazzi, 2024, https://doi.org/10.1029/2023wr036589), a limitation that PINOs can partially mitigate by incorporating data alongside the physics loss. We evaluate the method on experiments of increasing physical complexity: a radial dam break, constant BCs with and without friction, time‐dependent BCs, and a real‐world test case. The results show that PINOs capture key flood dynamics, particularly water depth, while reducing inference time by up to two orders of magnitude compared to numerical solvers. Relative test errors for water depth ranged from 0.3% for the radial dam break to 10.9% for cases with bottom topography, 7.3% with friction, and 9.0% under time‐dependent BCs. For the real‐world case, an additional data loss term yielded a water depth error of 25.8%. Larger errors were consistently observed for velocity components. The findings establish PINOs as a promising complement to traditional numerical solvers, offering a balance between computational efficiency and solution accuracy. Future work will focus on improving accuracy and extending the framework toward real‐world flood forecasting applications.

Forests are carbon sinks essential for climate change mitigation. However, increased harvests and natural disturbances across Europe have recently challenged this role. To project the future carbon sink capacity of Europe's forests, we integrated country reports from the United Nations Framework Convention on Climate Change with remote sensing maps of disturbances and above-ground biomass. Our model simulates biomass dynamics at 18 km resolution from 2010 to 2030, predicting a 39% decrease in the EU-27 forest carbon sink, driven by disturbances outpacing biomass recovery. Consequently, the 2030 forest carbon sink will fall 27% short of the estimated target consistent with EU-27 climate objectives. We demonstrate that the three billion trees initiative is insufficient for climate change mitigation and needs to be combined with a 28% reduction in forest harvests from 2025 to 2030 to meet these targets.

Monitoring canopy height change is essential for understanding carbon sinks and forest dynamics. Remote sensing enables consistent, large-scale observations of such changes, increasingly integrated with deep learning architectures such as Geospatial Foundation Models (GFMs). However, existing methods and datasets frame the problem as binary change detection, which overlooks both the continuous nature of change, especially for vegetation, and the inherent uncertainty in labels. We present the Canopy Height Change (CHC) dataset, providing 3 m resolution continuous canopy height differences and associated spatially resolved uncertainties across 10598 km^2 of northern and western Spain. The dataset is paired with a co-located time series of PlanetScope satellite imagery. Based on the dataset, we introduce the task of uncertainty-aware change regression, associated metrics and strategies for fine-tuning GFMs. Furthermore, we evaluate state-of-the-art GFMs and highlight promising directions and remaining challenges for advancing continuous canopy height change estimation.

Forest structure is an essential variable in forest management and conservation, as it has a direct impact on ecosystem processes and functions. Previous remote sensing studies have primarily focused on the vertical structure of forests, which requires laser point data and may not always be suited to distinguish plantations from old forests. Sub-meter resolution remote sensing data and tree crown segmentation techniques hold promise in offering detailed information that can support the characterization of forest structure from a horizontal perspective, offering new insights in the tree crown structure at scale. In this study, we generated a dataset with over 5 billion tree crowns and developed a Horizontal Structure Index (HSI) by analyzing spatial relationships among neighboring trees from remote sensing optical images. We first extracted the location and crown size of overstory trees from optical satellite and aerial imagery at sub-meter resolution. We subsequently calculated the distance between tree crown centers, their angles, the crown size and crown spacing, and linked this information with individual trees. We then used principal component analysis (PCA) to condense the structural information into the HSI and tested it in China, Rwanda and Denmark. Our result showed that the HSI has the potential to distinguish monoculture plantations from other forest types, which provides insights that extend beyond metrics derived from vertical forest structure. The proposed HSI is derived directly from tree-level attributes and supports a deeper understanding of forest structure from a horizontal perspective, complementing existing remote sensing-based metrics.

Abstract Computational visual intelligence has been shown to be able to comprehend the content of images, which has been widely used to foster a digitized society, but is often underutilized in applications related to the green transition and climate change mitigation. Here, we evaluate the capacity of convolutional neural networks (CNN) to interpret spatial semantic patterns in optical RGB images to directly estimate forest biomass, an essential climate parameter previously assessed from structural measures of trees. Trained with forest inventory plots, the CNN model demonstrates its learning via interpreting the composition of biomass at tree level, differing from traditional approaches reliant on conversions of aggregated parameters without explanatory rationale. The CNN approach yields consistently low bias across wide biomass ranges, whereas traditional models show insufficiency without information on tree height. Visually interpretable models link advanced computational tools with the power of data, facilitating the sustainable management of resources for a carbon-neutral society.

European wetlands store large carbon reserves1, but centuries of land use have eroded carbon stocks and biodiversity2. The European Union (EU) Nature Restoration Law (NRL)3 requires at least 30% of wetland ecosystems not in 'good condition' to be restored by 2030, yet spatially consistent information on wetland types and condition remains scarce. Using 10-m satellite imagery and machine learning, we map six seminatural open wetland types and land-use disturbance across 38 European countries. Wetlands are highly fragmented, with an estimated 27-33% of wetland area occurring in map-defined patches <25 ha and 7-11% in patches <1 ha, exposing many small sites missed by coarser products. We estimate that human activities affect 20.4 ± 3.4% of wetland areas (95% confidence interval), with inland wetland types most affected, and up to 5 Gt CO2-eq of soil carbon potentially lost relative to an undisturbed baseline. Translating disturbed area into NRL restoration targets, we find that several countries' pledges are broadly consistent with our 2030 estimates, whereas others lack quantified commitments despite substantial candidate areas identified by our maps. The resulting standardized, high-resolution products provide an EU-wide baseline tailored to the NRL and a reproducible template for linking satellite mapping to restoration targets, supporting progress tracking.

Although more than half of the urban population in sub-Saharan Africa reside in informal housing, knowledge about the material and spatial characteristics of such dwellings remains limited. This study examines informal housing practices in the Mabibo neighborhood of Dar es Salaam, Tanzania, using detailed surveys of the built environment and household data collection to investigate how dwellings are adapted to contextual conditions. Sociocultural adaptation strategies include spatial configurations to accommodate complex household structures, where extended families and tenants often reside in separate sub-units within the same plot, driving emergence of new courtyard housing typologies typified by built environment densification and increasing rental accommodation. Economic adaptation strategies include integration of income-generating functions in dwellings such as shops and rental units, incremental dwelling expansion based on available resources, and subdivision of plots to accommodate additional households. The study documents evolving construction practices with traditional housing being replaced by dwellings built from industrial materials and repurposed waste. Environmental adaptation strategies include raised floors to mitigate flooding, window screening to reduce exposure to mosquito-borne diseases, and suspended ceilings reported by residents as improving indoor thermal comfort. These findings show how peri-urban informal housing practices in rapidly urbanizing African cities evolve through adaptation to contextual conditions, such as land scarcity, constrained household economies, and availability of industrial construction materials linked to global supply chains.